2 Copyright (C) 2012 Carl Hetherington <cth@carlh.net>
4 This program is free software; you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation; either version 2 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program; if not, write to the Free Software
16 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 /** @file src/image.cc
21 * @brief A class to describe a video image.
26 #include <libswscale/swscale.h>
27 #include <libavutil/pixfmt.h>
28 #include <libavutil/pixdesc.h>
31 #include "exceptions.h"
33 #include "md5_digester.h"
41 using boost::shared_ptr;
45 Image::line_factor (int n) const
51 AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format);
53 throw PixelFormatError ("lines()", _pixel_format);
56 return pow (2.0f, d->log2_chroma_h);
59 /** @param n Component index.
60 * @return Number of lines in the image for the given component.
63 Image::lines (int n) const
65 return rint (ceil (static_cast<double>(size().height) / line_factor (n)));
68 /** @return Number of components */
70 Image::components () const
72 AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format);
74 throw PixelFormatError ("components()", _pixel_format);
77 if ((d->flags & PIX_FMT_PLANAR) == 0) {
81 return d->nb_components;
84 /** Crop this image, scale it to `inter_size' and then place it in a black frame of `out_size' */
86 Image::crop_scale_window (Crop crop, libdcp::Size inter_size, libdcp::Size out_size, Scaler const * scaler, AVPixelFormat out_format, bool out_aligned) const
89 /* Empirical testing suggests that sws_scale() will crash if
90 the input image is not aligned.
94 assert (out_size.width >= inter_size.width);
95 assert (out_size.height >= inter_size.height);
97 /* Here's an image of out_size */
98 shared_ptr<Image> out (new Image (out_format, out_size, out_aligned));
101 /* Size of the image after any crop */
102 libdcp::Size const cropped_size = crop.apply (size ());
104 /* Scale context for a scale from cropped_size to inter_size */
105 struct SwsContext* scale_context = sws_getContext (
106 cropped_size.width, cropped_size.height, pixel_format(),
107 inter_size.width, inter_size.height, out_format,
108 scaler->ffmpeg_id (), 0, 0, 0
111 if (!scale_context) {
112 throw StringError (N_("Could not allocate SwsContext"));
115 /* Prepare input data pointers with crop */
116 uint8_t* scale_in_data[components()];
117 for (int c = 0; c < components(); ++c) {
118 scale_in_data[c] = data()[c] + int (rint (bytes_per_pixel(c) * crop.left)) + stride()[c] * (crop.top / line_factor(c));
121 /* Corner of the image within out_size */
122 Position<int> const corner ((out_size.width - inter_size.width) / 2, (out_size.height - inter_size.height) / 2);
124 uint8_t* scale_out_data[out->components()];
125 for (int c = 0; c < out->components(); ++c) {
126 scale_out_data[c] = out->data()[c] + int (rint (out->bytes_per_pixel(c) * corner.x)) + out->stride()[c] * corner.y;
131 scale_in_data, stride(),
132 0, cropped_size.height,
133 scale_out_data, out->stride()
136 sws_freeContext (scale_context);
142 Image::scale (libdcp::Size out_size, Scaler const * scaler, AVPixelFormat out_format, bool out_aligned) const
145 /* Empirical testing suggests that sws_scale() will crash if
146 the input image is not aligned.
150 shared_ptr<Image> scaled (new Image (out_format, out_size, out_aligned));
152 struct SwsContext* scale_context = sws_getContext (
153 size().width, size().height, pixel_format(),
154 out_size.width, out_size.height, out_format,
155 scaler->ffmpeg_id (), 0, 0, 0
162 scaled->data(), scaled->stride()
165 sws_freeContext (scale_context);
171 Image::crop (Crop crop, bool aligned) const
173 libdcp::Size cropped_size = crop.apply (size ());
174 shared_ptr<Image> out (new Image (pixel_format(), cropped_size, aligned));
176 for (int c = 0; c < components(); ++c) {
177 int const crop_left_in_bytes = bytes_per_pixel(c) * crop.left;
178 /* bytes_per_pixel() could be a fraction; in this case the stride will be rounded
179 up, and we need to make sure that we copy over the width (up to the stride)
180 rather than short of the width; hence the ceil() here.
182 int const cropped_width_in_bytes = ceil (bytes_per_pixel(c) * cropped_size.width);
184 /* Start of the source line, cropped from the top but not the left */
185 uint8_t* in_p = data()[c] + (crop.top / out->line_factor(c)) * stride()[c];
186 uint8_t* out_p = out->data()[c];
188 for (int y = 0; y < out->lines(c); ++y) {
189 memcpy (out_p, in_p + crop_left_in_bytes, cropped_width_in_bytes);
191 out_p += out->stride()[c];
198 /** Blacken a YUV image whose bits per pixel is rounded up to 16 */
200 Image::yuv_16_black (uint16_t v, bool alpha)
202 memset (data()[0], 0, lines(0) * stride()[0]);
203 for (int i = 1; i < 3; ++i) {
204 int16_t* p = reinterpret_cast<int16_t*> (data()[i]);
205 for (int y = 0; y < lines(i); ++y) {
206 /* We divide by 2 here because we are writing 2 bytes at a time */
207 for (int x = 0; x < line_size()[i] / 2; ++x) {
210 p += stride()[i] / 2;
215 memset (data()[3], 0, lines(3) * stride()[3]);
220 Image::swap_16 (uint16_t v)
222 return ((v >> 8) & 0xff) | ((v & 0xff) << 8);
228 /* U/V black value for 8-bit colour */
229 static uint8_t const eight_bit_uv = (1 << 7) - 1;
230 /* U/V black value for 9-bit colour */
231 static uint16_t const nine_bit_uv = (1 << 8) - 1;
232 /* U/V black value for 10-bit colour */
233 static uint16_t const ten_bit_uv = (1 << 9) - 1;
234 /* U/V black value for 16-bit colour */
235 static uint16_t const sixteen_bit_uv = (1 << 15) - 1;
237 switch (_pixel_format) {
238 case PIX_FMT_YUV420P:
239 case PIX_FMT_YUV422P:
240 case PIX_FMT_YUV444P:
241 case PIX_FMT_YUV411P:
242 memset (data()[0], 0, lines(0) * stride()[0]);
243 memset (data()[1], eight_bit_uv, lines(1) * stride()[1]);
244 memset (data()[2], eight_bit_uv, lines(2) * stride()[2]);
247 case PIX_FMT_YUVJ420P:
248 case PIX_FMT_YUVJ422P:
249 case PIX_FMT_YUVJ444P:
250 memset (data()[0], 0, lines(0) * stride()[0]);
251 memset (data()[1], eight_bit_uv + 1, lines(1) * stride()[1]);
252 memset (data()[2], eight_bit_uv + 1, lines(2) * stride()[2]);
255 case PIX_FMT_YUV422P9LE:
256 case PIX_FMT_YUV444P9LE:
257 yuv_16_black (nine_bit_uv, false);
260 case PIX_FMT_YUV422P9BE:
261 case PIX_FMT_YUV444P9BE:
262 yuv_16_black (swap_16 (nine_bit_uv), false);
265 case PIX_FMT_YUV422P10LE:
266 case PIX_FMT_YUV444P10LE:
267 yuv_16_black (ten_bit_uv, false);
270 case PIX_FMT_YUV422P16LE:
271 case PIX_FMT_YUV444P16LE:
272 yuv_16_black (sixteen_bit_uv, false);
275 case PIX_FMT_YUV444P10BE:
276 case PIX_FMT_YUV422P10BE:
277 yuv_16_black (swap_16 (ten_bit_uv), false);
280 case AV_PIX_FMT_YUVA420P9BE:
281 case AV_PIX_FMT_YUVA422P9BE:
282 case AV_PIX_FMT_YUVA444P9BE:
283 yuv_16_black (swap_16 (nine_bit_uv), true);
286 case AV_PIX_FMT_YUVA420P9LE:
287 case AV_PIX_FMT_YUVA422P9LE:
288 case AV_PIX_FMT_YUVA444P9LE:
289 yuv_16_black (nine_bit_uv, true);
292 case AV_PIX_FMT_YUVA420P10BE:
293 case AV_PIX_FMT_YUVA422P10BE:
294 case AV_PIX_FMT_YUVA444P10BE:
295 yuv_16_black (swap_16 (ten_bit_uv), true);
298 case AV_PIX_FMT_YUVA420P10LE:
299 case AV_PIX_FMT_YUVA422P10LE:
300 case AV_PIX_FMT_YUVA444P10LE:
301 yuv_16_black (ten_bit_uv, true);
304 case AV_PIX_FMT_YUVA420P16BE:
305 case AV_PIX_FMT_YUVA422P16BE:
306 case AV_PIX_FMT_YUVA444P16BE:
307 yuv_16_black (swap_16 (sixteen_bit_uv), true);
310 case AV_PIX_FMT_YUVA420P16LE:
311 case AV_PIX_FMT_YUVA422P16LE:
312 case AV_PIX_FMT_YUVA444P16LE:
313 yuv_16_black (sixteen_bit_uv, true);
321 case PIX_FMT_RGB555LE:
322 memset (data()[0], 0, lines(0) * stride()[0]);
325 case PIX_FMT_UYVY422:
327 int const Y = lines(0);
328 int const X = line_size()[0];
329 uint8_t* p = data()[0];
330 for (int y = 0; y < Y; ++y) {
331 for (int x = 0; x < X / 4; ++x) {
332 *p++ = eight_bit_uv; // Cb
334 *p++ = eight_bit_uv; // Cr
342 throw PixelFormatError ("make_black()", _pixel_format);
347 Image::alpha_blend (shared_ptr<const Image> other, Position<int> position)
349 /* Only implemented for RGBA onto RGB24 so far */
350 assert (_pixel_format == PIX_FMT_RGB24 && other->pixel_format() == PIX_FMT_RGBA);
352 int start_tx = position.x;
356 start_ox = -start_tx;
360 int start_ty = position.y;
364 start_oy = -start_ty;
368 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
369 uint8_t* tp = data()[0] + ty * stride()[0] + position.x * 3;
370 uint8_t* op = other->data()[0] + oy * other->stride()[0];
371 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
372 float const alpha = float (op[3]) / 255;
373 tp[0] = (tp[0] * (1 - alpha)) + op[0] * alpha;
374 tp[1] = (tp[1] * (1 - alpha)) + op[1] * alpha;
375 tp[2] = (tp[2] * (1 - alpha)) + op[2] * alpha;
383 Image::copy (shared_ptr<const Image> other, Position<int> position)
385 /* Only implemented for RGB24 onto RGB24 so far */
386 assert (_pixel_format == PIX_FMT_RGB24 && other->pixel_format() == PIX_FMT_RGB24);
387 assert (position.x >= 0 && position.y >= 0);
389 int const N = min (position.x + other->size().width, size().width) - position.x;
390 for (int ty = position.y, oy = 0; ty < size().height && oy < other->size().height; ++ty, ++oy) {
391 uint8_t * const tp = data()[0] + ty * stride()[0] + position.x * 3;
392 uint8_t * const op = other->data()[0] + oy * other->stride()[0];
393 memcpy (tp, op, N * 3);
398 Image::read_from_socket (shared_ptr<Socket> socket)
400 for (int i = 0; i < components(); ++i) {
401 uint8_t* p = data()[i];
402 for (int y = 0; y < lines(i); ++y) {
403 socket->read (p, line_size()[i]);
410 Image::write_to_socket (shared_ptr<Socket> socket) const
412 for (int i = 0; i < components(); ++i) {
413 uint8_t* p = data()[i];
414 for (int y = 0; y < lines(i); ++y) {
415 socket->write (p, line_size()[i]);
423 Image::bytes_per_pixel (int c) const
425 AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format);
427 throw PixelFormatError ("lines()", _pixel_format);
430 if (c >= components()) {
434 float bpp[4] = { 0, 0, 0, 0 };
436 bpp[0] = floor ((d->comp[0].depth_minus1 + 1 + 7) / 8);
437 if (d->nb_components > 1) {
438 bpp[1] = floor ((d->comp[1].depth_minus1 + 1 + 7) / 8) / pow (2.0f, d->log2_chroma_w);
440 if (d->nb_components > 2) {
441 bpp[2] = floor ((d->comp[2].depth_minus1 + 1 + 7) / 8) / pow (2.0f, d->log2_chroma_w);
443 if (d->nb_components > 3) {
444 bpp[3] = floor ((d->comp[3].depth_minus1 + 1 + 7) / 8) / pow (2.0f, d->log2_chroma_w);
447 if ((d->flags & PIX_FMT_PLANAR) == 0) {
448 /* Not planar; sum them up */
449 return bpp[0] + bpp[1] + bpp[2] + bpp[3];
455 /** Construct a Image of a given size and format, allocating memory
458 * @param p Pixel format.
459 * @param s Size in pixels.
461 Image::Image (AVPixelFormat p, libdcp::Size s, bool aligned)
472 _data = (uint8_t **) wrapped_av_malloc (4 * sizeof (uint8_t *));
473 _data[0] = _data[1] = _data[2] = _data[3] = 0;
475 _line_size = (int *) wrapped_av_malloc (4 * sizeof (int));
476 _line_size[0] = _line_size[1] = _line_size[2] = _line_size[3] = 0;
478 _stride = (int *) wrapped_av_malloc (4 * sizeof (int));
479 _stride[0] = _stride[1] = _stride[2] = _stride[3] = 0;
481 for (int i = 0; i < components(); ++i) {
482 _line_size[i] = ceil (_size.width * bytes_per_pixel(i));
483 _stride[i] = stride_round_up (i, _line_size, _aligned ? 32 : 1);
485 /* The assembler function ff_rgb24ToY_avx (in libswscale/x86/input.asm)
486 uses a 16-byte fetch to read three bytes (R/G/B) of image data.
487 Hence on the last pixel of the last line it reads over the end of
488 the actual data by 1 byte. If the width of an image is a multiple
489 of the stride alignment there will be no padding at the end of image lines.
490 OS X crashes on this illegal read, though other operating systems don't
491 seem to mind. The nasty + 1 in this malloc makes sure there is always a byte
492 for that instruction to read safely.
494 Further to the above, valgrind is now telling me that ff_rgb24ToY_ssse3
495 over-reads by more then _avx. I can't follow the code to work out how much,
496 so I'll just over-allocate by 32 bytes and have done with it. Empirical
497 testing suggests that it works.
499 _data[i] = (uint8_t *) wrapped_av_malloc (_stride[i] * lines (i) + 32);
503 Image::Image (Image const & other)
504 : libdcp::Image (other)
505 , _pixel_format (other._pixel_format)
506 , _aligned (other._aligned)
510 for (int i = 0; i < components(); ++i) {
511 uint8_t* p = _data[i];
512 uint8_t* q = other._data[i];
513 for (int j = 0; j < lines(i); ++j) {
514 memcpy (p, q, _line_size[i]);
516 q += other.stride()[i];
521 Image::Image (AVFrame* frame)
522 : libdcp::Image (libdcp::Size (frame->width, frame->height))
523 , _pixel_format (static_cast<AVPixelFormat> (frame->format))
528 for (int i = 0; i < components(); ++i) {
529 uint8_t* p = _data[i];
530 uint8_t* q = frame->data[i];
531 for (int j = 0; j < lines(i); ++j) {
532 memcpy (p, q, _line_size[i]);
534 /* AVFrame's linesize is what we call `stride' */
535 q += frame->linesize[i];
540 Image::Image (shared_ptr<const Image> other, bool aligned)
541 : libdcp::Image (other)
542 , _pixel_format (other->_pixel_format)
547 for (int i = 0; i < components(); ++i) {
548 assert(line_size()[i] == other->line_size()[i]);
549 uint8_t* p = _data[i];
550 uint8_t* q = other->data()[i];
551 for (int j = 0; j < lines(i); ++j) {
552 memcpy (p, q, line_size()[i]);
554 q += other->stride()[i];
560 Image::operator= (Image const & other)
562 if (this == &other) {
572 Image::swap (Image & other)
574 libdcp::Image::swap (other);
576 std::swap (_pixel_format, other._pixel_format);
578 for (int i = 0; i < 4; ++i) {
579 std::swap (_data[i], other._data[i]);
580 std::swap (_line_size[i], other._line_size[i]);
581 std::swap (_stride[i], other._stride[i]);
584 std::swap (_aligned, other._aligned);
587 /** Destroy a Image */
590 for (int i = 0; i < components(); ++i) {
595 av_free (_line_size);
606 Image::line_size () const
612 Image::stride () const
624 Image::aligned () const
630 Image::digest () const
632 MD5Digester digester;
634 for (int i = 0; i < components(); ++i) {
635 digester.add (data()[i], line_size()[i]);
638 return digester.get ();